Answering my own question sort of. I ended up building a process that submits each file and then submits the chat token. Also figured out that since it only accepts PDF's, it will only be able to extract text if that layer is still there, like a pdf created from a .docx, and if the text is embedded in an image it can't be scraped. You'll have to push it through an OCR solution first. Bedrock has that capability so I find it easier to accomplish a multi-doc analysis there vs the Azure OpenAI connector.